知识卡片
匹配粒度与返回粒度可分开
可用句向量匹配具体问题,再返回包含该句的较大chunk提供背景。这样兼顾精确定位和上下文,但需保存句到chunk映射并控制扩展长度;小单元并不总比大单元更好。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 表示学习与嵌入
原书出处 · 1 处
- · 查看原书依据
corresponding to the most similar vectors are then retrieved. This ensures that the embedding matching happens at a lower granularity, like a sentence, but the model receives the entirety of the chunk the sentence belongs to, thus providing sufficient background context to the model.
参考来源
- 位置:《Designing Large Language Model Applications》第11章《表示学习与嵌入》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch11.html)
- 结论依据:摘录中关于匹配粒度与返回粒度可分开的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:corresponding to the most similar vectors are then retrieved. This ensures that the embedding matching happens at a lower granularity, like a sentence, but the model receives the entirety of the chunk the sentence belongs to, thus providing sufficient background context to the model.